An AI-based greenhouse planting management and control system

Abstract

The present application provides an AI-based greenhouse planting management and control system, which relates to the field of intelligent planting management. It uses AI-based data analysis and processing algorithms to perform data regularization and time-series encoding on the collected greenhouse environmental parameter data, so as to automatically recommend the water inlet valve opening value at the next time point according to the cross-domain query matching features between the various greenhouse environmental parameters, and generate control instructions. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner, achieve precise control of irrigation and temperature control, and thus achieve more intelligent greenhouse planting management and control.

Description

Technical Field

[0001] The present application relates to the field of intelligent planting management, and more specifically, to an AI-based greenhouse planting management and control system. Background Art

[0002] The application of greenhouse cultivation technology effectively avoids the adverse effects of seasonal weather changes on crop growth by providing a controllable growth environment. The stability of this environment makes crop planting no longer restricted by seasons, realizing an uninterrupted production cycle throughout the year, thereby significantly increasing crop yields.

[0003] However, traditional greenhouse cultivation management and control methods mainly rely on farmers' experience and some basic automated equipment, such as timed irrigation systems or simple temperature control devices. Specifically, traditional greenhouse management methods lack real-time monitoring and precise control of environmental parameters. For example, the irrigation system may operate according to a fixed schedule without considering the actual soil moisture or weather conditions, which may lead to excessive or insufficient irrigation, wasting water resources and being detrimental to the healthy growth of crops. In addition, traditional management and control methods cannot effectively integrate and analyze multi-dimensional environmental data, making it impossible for managers to accurately assess the comprehensive environmental conditions in the greenhouse, making it difficult to achieve precise resource management and crop growth regulation, which not only limits the improvement of crop yield and quality, but may also lead to resource waste and increased environmental pressure.

[0004] Therefore, an optimized greenhouse planting management and control system is desired. Summary of the invention

[0005] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides an AI-based greenhouse planting management and control system, which uses an AI-based data analysis and processing algorithm to perform data regularization and time series encoding on the collected greenhouse environmental parameter data, so as to automatically recommend the water inlet valve opening value at the next time point according to the cross-domain query matching features between the various greenhouse environmental parameters, and generate control instructions. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner, achieve precise control of irrigation and temperature control, and thus achieve more intelligent greenhouse planting management and control.

[0006] According to one aspect of the present application, there is provided an AI-based greenhouse planting management and control system, comprising: a sensor group, an inlet valve and a drain valve, a valve motor for adjusting the inlet valve and the drain valve, part of the inlet valve and / or the drain valve comprising a 4G communication module, a LoRa gateway, and a greenhouse planting management and control cloud; the inlet valve and the drain valve are communicatively connected to the greenhouse planting management and control cloud through the 4G communication module or the LoRa gateway; the sensor group is communicatively connected to the greenhouse planting management and control cloud, the sensor group comprises an air temperature sensor, an air humidity sensor, a light intensity sensor, a CO2 concentration sensor and a soil moisture sensor; the valve motor is communicatively connected to the greenhouse planting management and control cloud, the greenhouse planting management and control cloud is used to issue control instructions to the valve motor; wherein the greenhouse planting management and control cloud comprises: a greenhouse environment parameter acquisition module, for acquiring a time series of greenhouse environment parameter data collected by the sensor group, the greenhouse environment parameter data comprising an air temperature value, an air humidity value, a light intensity value, a CO2 concentration value and a soil moisture value; a greenhouse environment parameter timing coding module A block is used to divide the time series of the greenhouse environmental parameter data according to the sample dimension and then perform data regularization according to the time dimension to obtain air temperature time series input vector, air humidity time series input vector, light intensity time series input vector, CO2 concentration time series input vector and soil moisture time series input vector, and then pass through a time encoder to obtain air temperature time series feature vector, air humidity time series feature vector, light intensity time series feature vector, CO2 concentration time series feature vector and soil moisture time series feature vector; a soil moisture query matching module is used to use the soil moisture time series feature vector as a query feature vector, and use the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector as a sequence of library feature vectors to construct an initial key matrix, and pass the query feature vector and the initial key matrix through a cross-domain attention gate query encoder based on key matrix local association enhancement to obtain a soil moisture cross-domain query matching feature vector; a control instruction generation module is used to obtain a recommended value of the water inlet valve opening at the next time point based on the soil moisture cross-domain query matching feature vector, and generate the control instruction.

[0007] In the above-mentioned AI-based greenhouse planting management and control system, the greenhouse environmental parameter time series encoding module includes: a greenhouse environmental parameter data structuring unit, which is used to divide the time series of the greenhouse environmental parameter data according to the sample dimension and then regularize the data according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector; a greenhouse environmental parameter data time series feature extraction unit, which is used to pass the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector through a time encoder based on an RNN model to obtain the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector, the CO2 concentration time series feature vector and the soil moisture time series feature vector.

[0008] In the above AI-based greenhouse planting management and control system, the soil moisture query matching module includes: an initial key matrix generation unit, which is used to perform linear transformation on each library feature vector in the sequence of the library feature vector to obtain an initial key matrix, wherein each row vector in the initial key matrix is ​​the library feature vector after each linear transformation; a soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient calculation unit, which is used to calculate the characteristic distribution difference energy coefficient between the soil moisture time series feature vector and each row vector in the initial key matrix to obtain a set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients; an update key matrix generation unit, which is used to compare each soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient in the set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients with a predetermined threshold, and in response to the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient being less than the predetermined threshold , using the row vector corresponding to the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient and the mean vector between the soil moisture time series feature vector as the updated row vector to obtain an updated key matrix; a soil moisture time series feature embedding unit, used to use the value embedding matrix to process the soil moisture time series feature vector to obtain a soil moisture time series value vector; a soil moisture-atmospheric environment parameter cross-domain query unit, used to use the soil moisture time series feature vector as a query vector, each updated row vector in the update key matrix as a key vector and the soil moisture time series value vector as an input into a cross-domain attention gate query module based on a Transformer structure to obtain a sequence of soil moisture-atmospheric environment parameter cross-domain query attention vectors; a soil moisture cross-domain query matching feature generation unit, used to calculate the positional mean vector of the sequence of the soil moisture-atmospheric environment parameter cross-domain query attention vectors to obtain the soil moisture cross-domain query matching feature vector.

[0009] In the above-mentioned AI-based greenhouse planting management and control system, the soil moisture-atmospheric environmental parameter characteristic distribution difference energy coefficient calculation unit includes: a soil moisture-atmospheric environmental parameter characteristic time series interaction subunit, which is used to calculate the time series interaction vector between the soil moisture time series characteristic vector and each row vector in the initial key matrix to obtain a set of soil moisture-atmospheric environmental parameter time series correlation interaction vectors; a characteristic distribution difference energy coefficient generation subunit, which is used to calculate the characteristic distribution difference energy coefficient of each soil moisture-atmospheric environmental parameter time series correlation interaction vector in the set of soil moisture-atmospheric environmental parameter time series correlation interaction vectors to obtain a set of soil moisture-atmospheric environmental parameter characteristic distribution difference energy coefficients.

[0010] In the above-mentioned AI-based greenhouse planting management and control system, the soil moisture-atmospheric environmental parameter characteristic time series interaction subunit is used to: respectively calculate the position difference, position point multiplication and position addition between the soil moisture time series characteristic vector and the row vector to obtain the soil moisture-environment difference vector, the soil moisture-environment point product vector and the soil moisture-environment sum vector; cascade the soil moisture-environment difference vector, the soil moisture-environment point product vector and the soil moisture-environment sum vector and perform one-dimensional convolution encoding to obtain a soil moisture-environment multi-dimensional fusion vector; perform local window-based maximum pooling processing on the soil moisture-environment multi-dimensional fusion vector to obtain the soil moisture-atmospheric environmental parameter time series correlation interaction vector.

[0011] In the above-mentioned AI-based greenhouse planting management and control system, the characteristic distribution difference energy coefficient generating subunit is used to: respectively extract the maximum value of the soil moisture-atmospheric environment parameter time series correlation interaction vector to obtain the soil moisture-atmospheric environment parameter time series correlation maximum value; respectively calculate the mean and variance of the soil moisture-atmospheric environment parameter time series correlation interaction vector to obtain the soil moisture-atmospheric environment parameter time series correlation mean and the soil moisture-atmospheric environment parameter time series correlation variance; add the soil moisture-atmospheric environment parameter time series correlation variance and hyperparameters to obtain the soil moisture-atmospheric environment parameter time series correlation first difference energy coefficient; calculate the soil moisture-atmospheric environment parameter time series correlation maximum value and ... the soil moisture-atmospheric environment parameter time series correlation variance and hyperparameters to obtain the soil moisture-atmospheric environment parameter time series correlation first difference energy coefficient; calculate the soil moisture-atmospheric environment parameter time series correlation maximum value and the soil moisture-atmospheric environment parameter time series correlation The square of the difference between the mean values ​​of the time series association of environmental parameters is obtained to obtain the soil moisture-atmospheric environmental parameter time series difference value; the soil moisture-atmospheric environmental parameter time series association variance is multiplied by constant two to obtain the modulated soil moisture-atmospheric environmental parameter time series association variance; the soil moisture-atmospheric environmental parameter time series difference value and the modulated soil moisture-atmospheric environmental parameter time series association variance are added to the hyperparameter to obtain the soil moisture-atmospheric environmental parameter time series association second difference energy coefficient; the soil moisture-atmospheric environmental parameter time series association first difference energy coefficient is divided by the soil moisture-atmospheric environmental parameter time series association second difference energy coefficient to obtain the soil moisture-atmospheric environmental parameter characteristic distribution difference energy coefficient.

[0012] In the above-mentioned AI-based greenhouse planting management and control system, the soil moisture time series feature embedding unit is used to: after matrix multiplication of the soil moisture time series feature vector and the value embedding matrix, the obtained soil moisture time series embedding matrix and the value bias vector are added by position to obtain the soil moisture time series value vector.

[0013] In the above-mentioned AI-based greenhouse planting management and control system, the soil moisture-atmospheric environment parameter cross-domain query unit is used to: calculate the product between the query vector and the transposed vector of the update row vector to obtain the query-update association matrix, and then divide the query-update association matrix by the square root of the length of the update row vector by position to obtain a query-update modulation matrix; input the query-update modulation matrix into the softmax function to obtain a query-update activation matrix; multiply the query-update activation matrix by the soil moisture time series value vector to obtain the soil moisture-atmospheric environment parameter cross-domain query attention vector.

[0014] In the above AI-based greenhouse planting management and control system, the control instruction generation module is used to: pass the soil moisture cross-domain query matching feature vector through the decoder-based water inlet valve controller to obtain the recommended value of the water inlet valve opening at the next time point; generate the control instruction based on the comparison between the current opening value of the water inlet valve and the recommended value of the water inlet valve opening at the next time point.

[0015] Compared with the prior art, the AI-based greenhouse planting management and control system provided by the present application uses AI-based data analysis and processing algorithms to perform data regularization and time-series encoding on the collected greenhouse environmental parameter data, so as to automatically recommend the water inlet valve opening value at the next time point according to the cross-domain query matching features between the various greenhouse environmental parameters, and generate control instructions. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner, achieve precise control of irrigation and temperature control, and thus achieve more intelligent greenhouse planting management and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] In the attached picture: Figure 1 It is a block diagram of an AI-based greenhouse planting management and control system according to an embodiment of the present application.

[0018] Figure 2 It is a block diagram of the greenhouse planting control cloud in the AI-based greenhouse planting control system according to an embodiment of the present application.

[0019] Figure 3 This is a schematic diagram of data flow in the greenhouse planting control cloud in the AI-based greenhouse planting control system according to an embodiment of the present application.

[0020] Figure 4 It is a block diagram of the greenhouse environmental parameter timing encoding module in the AI-based greenhouse planting management and control system according to an embodiment of the present application.

[0021] Figure 5 This is a block diagram of a soil moisture query and matching module in an AI-based greenhouse planting management and control system according to an embodiment of the present application.

[0022] Figure 6It is a block diagram of a soil moisture-atmospheric environmental parameter characteristic distribution difference energy coefficient calculation unit in an AI-based greenhouse planting management and control system according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] As population growth and living standards improve, the global demand for food continues to increase, driving innovation in agricultural technology to meet the growing demand for food security. Greenhouse cultivation technology overcomes the adverse effects of seasonal climate on crop growth by providing a stable and controllable growth environment, achieving continuous production throughout the year, and thus significantly increasing crop yields.

[0025] However, traditional greenhouse cultivation mainly relies on farmers' experience and basic automation equipment, such as timed irrigation and temperature control devices, and lacks real-time monitoring and precise control of environmental parameters. For example, the irrigation system may operate according to a fixed schedule rather than actual soil moisture or weather conditions, resulting in water waste and crop growth problems. In addition, traditional methods fail to effectively integrate multi-dimensional environmental data, making it difficult for managers to accurately assess the greenhouse environment, affecting the accuracy of resource management and crop regulation, limiting yield and quality improvements, and increasing resource waste and environmental pressure.

[0026] Based on this, this application proposes an AI-based greenhouse planting management and control system. Figure 1 FIG. 1 is a block diagram of an AI-based greenhouse planting management and control system according to an embodiment of the present application. Figure 1 As shown, in the AI-based greenhouse planting control system 100, it includes: a sensor group 110, a water inlet valve and a drain valve 120, a valve motor 130 for adjusting the water inlet valve and the drain valve, and some of the water inlet valves and / or the drain valves include a 4G communication module, a LoRa gateway, and a greenhouse planting control cloud 140. The water inlet valve and the drain valve are communicatively connected to the greenhouse planting control cloud through the 4G communication module or the LoRa gateway; the sensor group is communicatively connected to the greenhouse planting control cloud, and the sensor group includes an air temperature sensor, an air humidity sensor, a light intensity sensor, a CO2 concentration sensor, and a soil moisture sensor; the valve motor is communicatively connected to the greenhouse planting control cloud, and the greenhouse planting control cloud is used to send control instructions to the valve motor.

[0027] It should be understood that the sensor group consisting of air temperature sensor, air humidity sensor, light intensity sensor, CO2 concentration sensor and soil moisture sensor can monitor the environmental parameters in the greenhouse in real time, help understand the growth environment of greenhouse crops, and accurately evaluate the comprehensive environmental conditions in the greenhouse by performing time series analysis on these environmental parameter data, so as to achieve precise resource management and crop growth regulation. The inlet valve and drain valve in the greenhouse planting control system are used to adjust the irrigation and drainage process to ensure that greenhouse crops get the appropriate moisture. The valve motor is used to adjust the inlet valve and drain valve, and is responsible for adjusting the valve opening according to the control command to improve the efficiency and accuracy of irrigation and drainage. The 4G communication module and LoRa gateway are used to connect the inlet valve and drain valve to the greenhouse planting control cloud to support remote data transmission, so that the system can achieve remote monitoring and control, enhance flexibility and operability, especially in large-scale greenhouse management, can reduce dependence on on-site personnel. The greenhouse planting control cloud is responsible for receiving data from the sensor group, performing data analysis and feature extraction, and generating control instructions to send to the valve motor. This cloud-based centralized management and intelligent analysis capabilities enable the system to adapt to environmental changes in real time, optimize resource allocation, improve the growth environment and management efficiency of crops, and ultimately increase crop yield and quality.

[0028] Correspondingly, in the greenhouse planting control cloud, the time series of greenhouse environmental parameter data (air temperature value, air humidity value, light intensity value, CO2 concentration value and soil moisture value) is collected through the sensor group, and the AI-based data analysis and processing algorithm is used to perform data regularization and time series encoding, so as to automatically recommend the water inlet valve opening value at the next time point according to the cross-domain query matching characteristics between each greenhouse environmental parameter, and generate control instructions. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner, realize precise control of irrigation and temperature control, and thus realize more intelligent greenhouse planting control.

[0029] Figure 2 It is a block diagram of the greenhouse planting control cloud in the AI-based greenhouse planting control system according to an embodiment of the present application. Figure 3 Schematic diagram of data flow in the greenhouse planting control cloud in the AI-based greenhouse planting control system according to an embodiment of the present application. Figure 2 and Figure 3As shown, in the greenhouse planting management and control cloud 140, it includes: a greenhouse environment parameter acquisition module 141, which is used to obtain the time series of greenhouse environment parameter data collected by the sensor group, and the greenhouse environment parameter data includes air temperature value, air humidity value, light intensity value, CO2 concentration value and soil humidity value; a greenhouse environment parameter time series encoding module 142, which is used to divide the time series of the greenhouse environment parameter data according to the sample dimension and then regularize the data according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil humidity time series input vector, and then pass the time encoder to obtain the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil humidity time series input vector. A soil moisture query matching module 143 is used to construct an initial key matrix using the soil moisture time series feature vector as a query feature vector, and using the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector as a sequence of library feature vectors, and to obtain a soil moisture cross-domain query matching feature vector by passing the query feature vector and the initial key matrix through a cross-domain attention gate query encoder based on key matrix local association enhancement; a control instruction generation module 144 is used to obtain a recommended value of the water inlet valve opening at the next time point based on the soil moisture cross-domain query matching feature vector, and to generate the control instruction.

[0030] In the embodiment of the present application, the greenhouse environment parameter acquisition module 141 is used to obtain the time series of greenhouse environment parameter data collected by the sensor group, and the greenhouse environment parameter data includes air temperature value, air humidity value, light intensity value, CO2 concentration value and soil moisture value. It should be understood that the air temperature value affects the photosynthesis, respiration and growth rate of greenhouse crops; the air humidity value affects the transpiration and water absorption of greenhouse crops; the light intensity value directly affects the photosynthesis efficiency of greenhouse crops; CO2 is the raw material for photosynthesis, and the appropriate CO2 concentration value helps to increase the growth rate and yield of crops; the soil moisture value directly affects the water supply of greenhouse crops. In general, these data provide the real-time status of the environment in the greenhouse. By analyzing these data, it can help identify environmental change trends and patterns, thereby predicting future environmental changes and making corresponding valve opening adjustments to optimize irrigation strategies, thereby improving crop growth efficiency and yield.

[0031] In the embodiment of the present application, the greenhouse environment parameter time series encoding module 142 is used to divide the time series of the greenhouse environment parameter data according to the sample dimension and then perform data regularization according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil humidity time series input vector, and then pass through the time encoder to obtain the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector, the CO2 concentration time series feature vector and the soil humidity time series feature vector. Specifically, Figure 4 FIG. 1 is a block diagram of a greenhouse environment parameter timing coding module in an AI-based greenhouse planting management and control system according to an embodiment of the present application. Figure 4 As shown, the greenhouse environment parameter time series encoding module 142 includes: a greenhouse environment parameter data structuring unit 1421, which is used to divide the time series of the greenhouse environment parameter data according to the sample dimension and then regularize the data according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector; a greenhouse environment parameter data time series feature extraction unit 1422, which is used to pass the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector through a time encoder based on an RNN model to obtain the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector, the CO2 concentration time series feature vector and the soil moisture time series feature vector.

[0032] It should be understood that each greenhouse environmental parameter of the greenhouse environmental parameter data has its own characteristic information, and each parameter has characteristic information in the time dimension, such as fluctuations and periodic changes in a short period of time. Therefore, in order to analyze the characteristics of each environmental parameter more carefully, in the technical solution of the present application, the time series of the greenhouse environmental parameter data is divided according to the sample dimension and then the data is regularized according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector. Further, in order to better capture and mine the time series characteristics of each greenhouse environmental parameter in the time dimension, in the technical solution of the present application, the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector are passed through a time encoder based on an RNN model to obtain an air temperature time series feature vector, an air humidity time series feature vector, a light intensity time series feature vector, a CO2 concentration time series feature vector and a soil moisture time series feature vector. Those skilled in the art should know that the RNN model is a neural network architecture that specializes in processing sequence data. It can capture the time-dependent information in the input sequence data through a recurrent connection mechanism. By processing the time series input vector data of multiple greenhouse environmental parameters through a time encoder based on the RNN model, it can not only effectively extract and represent the time series characteristics of the greenhouse environmental parameters, but also better predict future greenhouse environmental changes, thereby improving the accuracy of system decision-making.

[0033] In an embodiment of the present application, the soil moisture query matching module 143 is used to construct an initial key matrix using the soil moisture time series feature vector as a query feature vector, and using the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector as a sequence of library feature vectors, and pass the query feature vector and the initial key matrix through a cross-domain attention gate query encoder based on local association enhancement of the key matrix to obtain a soil moisture cross-domain query matching feature vector. Accordingly, considering that in a greenhouse environment, there are complex associations and mutual influences between various greenhouse environmental parameters in the time dimension, and in order to effectively capture the mutual influences between these parameters, so as to extract more representative features, in order to more comprehensively understand the interactions between environmental parameters, in the technical solution of the present application, a cross-domain attention gate query encoder based on local association enhancement of the key matrix is ​​introduced to use the soil moisture time series feature vector as the query feature vector, and the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector as a sequence of library feature vectors to construct an initial key matrix for query matching based on the query feature vector and the initial key matrix, thereby obtaining a cross-domain query matching feature vector of soil moisture. That is, the cross-domain attention gate query encoder based on local association enhancement of the key matrix adopts a cross-domain attention mechanism based on the Transformer architecture, and calculates self-attention by combining the query vector, key vector and value vector to achieve effective fusion of different features and information integration.

[0034] In detail, first, the sequence of the library feature vectors is linearly transformed to adjust the dimension between each feature to obtain the library feature vector after linear transformation, and it is constructed into an initial key matrix. Then, the feature distribution difference between each row vector in each initial key matrix and the soil moisture time series feature vector is calculated to obtain a set of soil moisture-atmospheric environmental parameter feature distribution difference energy coefficients. In particular, the feature distribution difference energy coefficient can identify which environmental parameters have significant differences in feature distribution with soil moisture, that is, since the soil moisture time series feature vector is a reference benchmark, the overall set of feature distribution difference energy coefficients also has a local time series correlation influence relationship between each row vector of the key matrix. Next, the feature distribution difference energy coefficient is compared with a predetermined threshold. For the case where it is less than the threshold, the mean vector of the query feature vector and the corresponding row vector is used to update the row vector in the key matrix to further strengthen the time series correlation between them and the query feature vector, and obtain an updated key matrix composed of multiple updated row vectors. Subsequently, the soil moisture time series feature vector is linearly transformed to obtain a value vector, and each row vector in the update key matrix is ​​used as a key vector. The query vector, value vector and key vector are input into the cross-domain attention gate query module based on the Transformer structure for local time series query matching semantic association encoding to effectively integrate information from different environmental parameters (air temperature, air humidity, light intensity and CO2 concentration), and realize cross-domain information fusion to obtain a sequence of soil moisture-atmospheric environmental parameter cross-domain query attention vectors. Finally, the sequence of query attention vectors is averaged to generate a comprehensive soil moisture cross-domain query matching feature vector that contains each local time series interaction.

[0035] Specifically, Figure 5 FIG. 1 is a block diagram of a soil moisture query matching module in an AI-based greenhouse planting management and control system according to an embodiment of the present application. Figure 5As shown, the soil moisture query matching module 143 includes: an initial key matrix generating unit 1431, which is used to perform linear transformation on each library feature vector in the sequence of the library feature vector to obtain an initial key matrix, wherein each row vector in the initial key matrix is ​​the library feature vector after each linear transformation; a soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient calculating unit 1432, which is used to calculate the characteristic distribution difference energy coefficient between the soil moisture time series feature vector and each row vector in the initial key matrix to obtain a set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients; an updating key matrix generating unit 1433, which is used to compare each soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient in the set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients with a predetermined threshold value, and in response to the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient being less than the predetermined threshold value, the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient is used to obtain a set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients. The row vector corresponding to the humidity-atmospheric environment parameter characteristic distribution difference energy coefficient and the mean vector between the soil moisture time series feature vector are used as the updated row vector to obtain the updated key matrix; the soil moisture time series feature embedding unit 1434 is used to process the soil moisture time series feature vector using the value embedding matrix to obtain the soil moisture time series value vector; the soil moisture-atmospheric environment parameter cross-domain query unit 1435 is used to input the soil moisture time series feature vector as the query vector, each updated row vector in the update key matrix as the key vector and the soil moisture time series value vector into the cross-domain attention gate query module based on the Transformer structure to obtain the sequence of soil moisture-atmospheric environment parameter cross-domain query attention vector; the soil moisture cross-domain query matching feature generation unit 1436 is used to calculate the positional mean vector of the sequence of the soil moisture-atmospheric environment parameter cross-domain query attention vector to obtain the soil moisture cross-domain query matching feature vector.

[0036] More specifically, Figure 6 FIG. 1 is a block diagram of a soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient calculation unit in an AI-based greenhouse planting management and control system according to an embodiment of the present application. Figure 6As shown, the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient calculation unit 1432 includes: a soil moisture-atmospheric environment parameter characteristic time series interaction subunit 14321, which is used to calculate the time series interaction vector between the soil moisture time series characteristic vector and each row vector in the initial key matrix to obtain a set of soil moisture-atmospheric environment parameter time series correlation interaction vectors; a characteristic distribution difference energy coefficient generation subunit 14322, which is used to calculate the characteristic distribution difference energy coefficient of each soil moisture-atmospheric environment parameter time series correlation interaction vector in the set of soil moisture-atmospheric environment parameter time series correlation interaction vectors to obtain the set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients.

[0037] More specifically, in an embodiment of the present application, the soil moisture-atmospheric environmental parameter characteristic time series interaction subunit is used to: respectively calculate the positional difference, positional dot product and positional addition between the soil moisture time series characteristic vector and the row vector to obtain a soil moisture-environment difference vector, a soil moisture-environment dot product vector and a soil moisture-environment sum vector; cascade the soil moisture-environment difference vector, the soil moisture-environment dot product vector and the soil moisture-environment sum vector and perform one-dimensional convolution encoding to obtain a soil moisture-environment multi-dimensional fusion vector; perform local window-based maximum pooling processing on the soil moisture-environment multi-dimensional fusion vector to obtain the soil moisture-atmospheric environmental parameter time series correlation interaction vector.

[0038] More specifically, in an embodiment of the present application, the characteristic distribution difference energy coefficient generating subunit is used to: respectively extract the maximum value of the soil moisture-atmospheric environment parameter time series correlation interaction vector to obtain the soil moisture-atmospheric environment parameter time series correlation maximum value; respectively calculate the mean and variance of the soil moisture-atmospheric environment parameter time series correlation interaction vector to obtain the soil moisture-atmospheric environment parameter time series correlation mean and the soil moisture-atmospheric environment parameter time series correlation variance; add the soil moisture-atmospheric environment parameter time series correlation variance and hyperparameters to obtain the soil moisture-atmospheric environment parameter time series correlation first difference energy coefficient; calculate the soil moisture-atmospheric environment parameter time series correlation maximum value and ... the soil moisture-atmospheric environment parameter time series correlation variance and hyperpara The square of the difference between the parameter time series association means is used to obtain the soil moisture-atmospheric environment parameter time series difference value; the soil moisture-atmospheric environment parameter time series association variance is multiplied by constant two to obtain the modulated soil moisture-atmospheric environment parameter time series association variance; the soil moisture-atmospheric environment parameter time series difference value and the modulated soil moisture-atmospheric environment parameter time series association variance are added to the hyperparameter to obtain the soil moisture-atmospheric environment parameter time series association second difference energy coefficient; the soil moisture-atmospheric environment parameter time series association first difference energy coefficient is divided by the soil moisture-atmospheric environment parameter time series association second difference energy coefficient to obtain the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient.

[0039] More specifically, in an embodiment of the present application, the soil moisture time series feature embedding unit is used to: after matrix multiplication of the soil moisture time series feature vector and the value embedding matrix, add the obtained soil moisture time series embedding matrix and the value bias vector by position to obtain the soil moisture time series value vector.

[0040] More specifically, in an embodiment of the present application, the soil moisture-atmospheric environment parameter cross-domain query unit is used to: calculate the product between the query vector and the transposed vector of the update row vector to obtain a query-update association matrix, and then divide the query-update association matrix by the square root of the length of the update row vector by position to obtain a query-update modulation matrix; input the query-update modulation matrix into a softmax function to obtain a query-update activation matrix; multiply the query-update activation matrix by the soil moisture time series value vector to obtain the soil moisture-atmospheric environment parameter cross-domain query attention vector.

[0041] In the embodiment of the present application, specifically, the soil moisture query matching module 143 is used to: use the soil moisture time series feature vector as the query feature vector, use the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector as a sequence of library feature vectors to construct an initial key matrix, and process the query feature vector and the initial key matrix through a cross-domain attention gate query encoder based on key matrix local association enhancement, and use the following cross-domain gated query to obtain the soil moisture cross-domain query matching feature vector; wherein the cross-domain gated query formula is: ;in, represents a sequence of library feature vectors, is the sequence of the library feature vector library feature vectors, is the initial bond matrix, is the first row vectors, The sequence of the library feature vector library feature vectors, is the weight matrix, is the bias vector, is the first row vectors, For Perform linear transformation operations, is the soil moisture time series feature vector, , and They are position difference, position dot multiplication and position addition respectively. For cascade processing, is a one-dimensional convolutional coding operation, is the maximum pooling operation, It is the first in the set of soil moisture-atmospheric environmental parameter time series correlation interaction vectors. Soil moisture-atmospheric environmental parameters time series correlation interaction vector, To extract the maximum value of a vector, and They are The mean and variance of is a hyperparameter, It is the first in the set of soil moisture-atmospheric environmental parameter characteristic distribution difference energy coefficients. Soil moisture-atmospheric environmental parameter characteristic distribution difference energy coefficient, is a predetermined threshold, and To update the key matrix update row vectors, and are the value embedding matrix and value bias vector respectively, is the soil moisture time series value vector, for The transposed vector of is the length of each update row vector, yes function, is the first in the sequence of the soil moisture-atmospheric environment parameter cross-domain query attention vector Soil moisture-atmospheric environment parameter cross-domain query attention vector, is the number of vectors in the sequence of the soil moisture-atmospheric environment parameter cross-domain query focus vector, is the soil moisture cross-domain query matching feature vector.

[0042] In the embodiment of the present application, the control instruction generation module 144 is used to obtain the recommended value of the water inlet valve opening at the next time point based on the soil moisture cross-domain query matching feature vector, and generate the control instruction. Specifically, in the embodiment of the present application, the control instruction generation module is used to: pass the soil moisture cross-domain query matching feature vector through the decoder-based water inlet valve controller to obtain the recommended value of the water inlet valve opening at the next time point; based on the comparison between the current opening value of the water inlet valve and the recommended value of the water inlet valve opening at the next time point, generate the control instruction. That is, the soil moisture cross-domain query matching feature vector obtained by cross-domain query matching using the soil moisture time series feature vector, the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector is decoded to automatically recommend the water inlet valve opening value at the next time point and generate the control instruction. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner and achieve precise control of irrigation and temperature control, thereby realizing more intelligent greenhouse planting management and control.

[0043] In particular, in the preferred example, considering that the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector, the CO2 concentration time series feature vector and the soil moisture time series feature vector respectively represent the cyclic time series correlation characteristics of the air temperature value, the air humidity value, the light intensity value, the CO2 concentration value and the soil moisture value, when performing cross-domain attention gate query encoding based on the query feature vector and the initial key matrix based on the key matrix local correlation enhancement, the key matrix association enhancement cross-domain attention difference under the feature time series correlation distribution difference caused by the source data time series distribution difference will make the soil moisture cross-domain query matching feature vector have a diversified set expression distribution of fusion features. Therefore, it is expected to improve the balance between the regression mapping accuracy and completeness of the soil moisture cross-domain query matching feature vector when it is decoded and regressed by the decoder-based water inlet valve controller, thereby improving the accuracy of the recommended value of the water inlet valve opening at the next time point.

[0044] In a preferred example, passing the soil moisture cross-domain query matching feature vector through the decoder-based water inlet valve controller to obtain the recommended value of the water inlet valve opening at the next time point includes: determining the soil moisture cross-domain query matching maximum eigenvalue and the soil moisture cross-domain query matching minimum eigenvalue of the soil moisture cross-domain query matching feature vector, and calculating the soil moisture cross-domain query matching mean and the soil moisture cross-domain query matching standard deviation of the feature set of the soil moisture cross-domain query matching feature vector; calculating the quotient of the soil moisture cross-domain query matching mean and the soil moisture cross-domain query matching standard deviation to obtain the soil moisture cross-domain query matching statistical normalization value; calculating the inverse of each eigenvalue of the soil moisture cross-domain query matching feature vector, and comparing it with the soil moisture cross-domain query matching maximum eigenvalue and After multiplying the difference of the minimum eigenvalues ​​of the soil moisture cross-domain query matching, the soil moisture cross-domain query matching statistical normalized value is subtracted to obtain a soil moisture cross-domain query matching distribution approximation vector; an exponential function with a natural constant as the base and each eigenvalue of the soil moisture cross-domain query matching distribution approximation vector is calculated, and the exponential function is subtracted from the soil moisture cross-domain query matching statistical normalized value to obtain a soil moisture cross-domain query matching probability approximation vector; the logarithm with base two of the absolute value of each eigenvalue of the soil moisture cross-domain query matching probability approximation vector is calculated to obtain an optimized soil moisture cross-domain query matching feature vector; the optimized soil moisture cross-domain query matching feature vector is passed through the decoder-based water inlet valve controller to obtain a recommended value of the water inlet valve opening at the next time point.

[0045] The optimized representation of the soil moisture cross-domain query matching feature vector is as follows: ;in, The cross-domain query matching feature vector for the soil moisture, , , and They are the feature sets of the soil moisture cross-domain query matching feature vectors The mean, standard deviation, minimum and maximum values ​​of is the logarithm to base 2, Indicates adding by position point. represents the natural exponential function, It means point multiplication by position. To calculate the inverse of each eigenvalue of the soil moisture cross-domain query matching eigenvector, Indicates the spread by position, Represents the optimized soil moisture cross-domain query matching feature vector.

[0046] Therefore, by performing random statistical normalization on the diversified feature set of the soil moisture cross-domain query matching feature vector, the confidence space constructed based on the overall eigenvalue of the soil moisture cross-domain query matching feature vector is approximated by the response hypothesis test of each eigenvalue of the soil moisture cross-domain query matching feature vector, thereby establishing the target reachability of the diversified feature distribution of the soil moisture cross-domain query matching feature vector to the unified regression target, so as to achieve the balanced executableness between mapping accuracy and mapping completeness in the decoding regression process based on the diversified feature distribution of the soil moisture cross-domain query matching feature vector, and improve the accuracy of the recommended value of the water inlet valve opening at the next time point obtained by the soil moisture cross-domain query matching feature vector through the decoder-based water inlet valve controller. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner, realize precise control of irrigation and temperature control, and thus realize more intelligent greenhouse planting management and control.

[0047] In summary, the AI-based greenhouse planting management and control system 100 based on the embodiment of the present application is explained, which uses AI-based data analysis and processing algorithms to perform data regularization and time series encoding on the collected greenhouse environmental parameter data, so as to automatically recommend the water inlet valve opening value at the next time point according to the cross-domain query matching characteristics between the various greenhouse environmental parameters, and generate control instructions. In this way, multiple key environmental parameters in the greenhouse can be monitored in real time, providing more comprehensive and accurate environmental status information. At the same time, it can respond to environmental changes in the greenhouse in a timely manner, achieve precise control of irrigation and temperature control, and thus achieve more intelligent greenhouse planting management and control.

[0048] As described above, the AI-based greenhouse planting management and control system 100 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for AI-based greenhouse planting management and control. In one example, the AI-based greenhouse planting management and control system 100 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the AI-based greenhouse planting management and control system 100 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the AI-based greenhouse planting management and control system 100 can also be one of the many hardware modules of the terminal device.

[0049] Alternatively, in another example, the AI-based greenhouse planting management and control system 100 and the terminal device may also be separate devices, and the AI-based greenhouse planting management and control system 100 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0050] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details of the above application are only for the purpose of illustration and ease of understanding, rather than limitation, and the above details do not limit the present application to the use of the above specific details to be implemented.

Claims

1. An AI-based greenhouse planting management and control system, characterized in that: include: A sensor group, an inlet valve and a drain valve, a valve motor for adjusting the inlet valve and the drain valve, part of the inlet valve and / or the drain valve includes a 4G communication module, a LoRa gateway, and a greenhouse planting control cloud; the inlet valve and the drain valve are communicatively connected to the greenhouse planting control cloud through the 4G communication module or the LoRa gateway; the sensor group is communicatively connected to the greenhouse planting control cloud, the sensor group includes an air temperature sensor, an air humidity sensor, a light intensity sensor, a CO2 concentration sensor and a soil moisture sensor; the valve motor is communicatively connected to the greenhouse planting control cloud, and the greenhouse planting control cloud is used to issue control instructions to the valve motor; wherein the greenhouse planting control cloud includes: a greenhouse environment parameter acquisition module, used to acquire the time series of greenhouse environment parameter data collected by the sensor group, the greenhouse environment parameter data including air temperature value, air humidity value, light intensity value, CO2 concentration value and soil moisture value; a greenhouse environment parameter timing encoding module, used to convert the time series of the greenhouse environment parameter data The time series is divided according to the sample dimension and then data is regularized according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil humidity time series input vector, and then the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector, the CO2 concentration time series feature vector and the soil humidity time series feature vector are obtained through the time encoder; a soil humidity query matching module is used to use the soil humidity time series feature vector as the query feature vector, and the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector and the CO2 concentration time series feature vector as the sequence of library feature vectors to construct an initial key matrix, and the query feature vector and the initial key matrix are passed through a cross-domain attention gate query encoder based on the local association enhancement of the key matrix to obtain a soil humidity cross-domain query matching feature vector; a control instruction generation module is used to obtain a recommended value of the water inlet valve opening at the next time point based on the soil humidity cross-domain query matching feature vector, and generate the control instruction; Wherein, the soil moisture query matching module includes: an initial key matrix generating unit, used for performing linear transformation on each library feature vector in the sequence of the library feature vector to obtain an initial key matrix, wherein each row vector in the initial key matrix is ​​the library feature vector after each linear transformation; a soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient calculating unit, used for calculating the characteristic distribution difference energy coefficient between the soil moisture time series feature vector and each row vector in the initial key matrix to obtain a set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients; an updating key matrix generating unit, used for comparing each soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient in the set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients with a predetermined threshold value, and in response to the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient being less than the predetermined threshold value, using the soil moisture as the query matching module; The row vector corresponding to the energy coefficient of the difference in the characteristic distribution of the degree-atmospheric environment parameter and the mean vector between the soil moisture time series feature vector are used as the updated row vector to obtain the updated key matrix; a soil moisture time series feature embedding unit is used to process the soil moisture time series feature vector using the value embedding matrix to obtain the soil moisture time series value vector; a soil moisture-atmospheric environment parameter cross-domain query unit is used to input the soil moisture time series feature vector as the query vector, each updated row vector in the update key matrix as the key vector and the soil moisture time series value vector into a cross-domain attention gate query module based on the Transformer structure to obtain a sequence of soil moisture-atmospheric environment parameter cross-domain query attention vectors; a soil moisture cross-domain query matching feature generation unit is used to calculate the positional mean vector of the sequence of the soil moisture-atmospheric environment parameter cross-domain query attention vectors to obtain the soil moisture cross-domain query matching feature vector; Wherein, the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient calculation unit includes: a soil moisture-atmospheric environment parameter characteristic time series interaction subunit, which is used to calculate the time series interaction vector between the soil moisture time series characteristic vector and each row vector in the initial key matrix to obtain a set of soil moisture-atmospheric environment parameter time series correlation interaction vectors; a characteristic distribution difference energy coefficient generation subunit, which is used to calculate the characteristic distribution difference energy coefficient of each soil moisture-atmospheric environment parameter time series correlation interaction vector in the set of soil moisture-atmospheric environment parameter time series correlation interaction vectors to obtain the set of soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficients; Among them, the soil moisture-atmospheric environment parameter characteristic time series interaction subunit is used to: respectively calculate the position difference, position point multiplication and position addition between the soil moisture time series characteristic vector and the row vector to obtain the soil moisture-environment difference vector, the soil moisture-environment point product vector and the soil moisture-environment sum vector; cascade the soil moisture-environment difference vector, the soil moisture-environment point product vector and the soil moisture-environment sum vector and perform one-dimensional convolution encoding to obtain a soil moisture-environment multi-dimensional fusion vector; perform local window-based maximum pooling processing on the soil moisture-environment multi-dimensional fusion vector to obtain the soil moisture-atmospheric environment parameter time series correlation interaction vector; Among them, the characteristic distribution difference energy coefficient generating subunit is used to: respectively extract the maximum value of the soil moisture-atmospheric environment parameter time series correlation interaction vector to obtain the soil moisture-atmospheric environment parameter time series correlation maximum value; respectively calculate the mean and variance of the soil moisture-atmospheric environment parameter time series correlation interaction vector to obtain the soil moisture-atmospheric environment parameter time series correlation mean and the soil moisture-atmospheric environment parameter time series correlation variance; add the soil moisture-atmospheric environment parameter time series correlation variance and hyperparameters to obtain the soil moisture-atmospheric environment parameter time series correlation first difference energy coefficient; calculate the soil moisture-atmospheric environment parameter time series correlation maximum value and the soil moisture-atmospheric environment parameter time series correlation The square of the difference in means is used to obtain the soil moisture-atmospheric environment parameter time series difference value; the soil moisture-atmospheric environment parameter time series correlation variance is multiplied by constant two to obtain the modulated soil moisture-atmospheric environment parameter time series correlation variance; the soil moisture-atmospheric environment parameter time series difference value and the modulated soil moisture-atmospheric environment parameter time series correlation variance are added to the hyperparameter to obtain the soil moisture-atmospheric environment parameter time series correlation second difference energy coefficient; the soil moisture-atmospheric environment parameter time series correlation first difference energy coefficient is divided by the soil moisture-atmospheric environment parameter time series correlation second difference energy coefficient to obtain the soil moisture-atmospheric environment parameter characteristic distribution difference energy coefficient.

2. The AI-based greenhouse planting management and control system according to claim 1 is characterized in that: The greenhouse environment parameter time series encoding module includes: a greenhouse environment parameter data structuring unit, which is used to divide the time series of the greenhouse environment parameter data according to the sample dimension and then regularize the data according to the time dimension to obtain the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector; a greenhouse environment parameter data time series feature extraction unit, which is used to pass the air temperature time series input vector, the air humidity time series input vector, the light intensity time series input vector, the CO2 concentration time series input vector and the soil moisture time series input vector through a time encoder based on an RNN model to obtain the air temperature time series feature vector, the air humidity time series feature vector, the light intensity time series feature vector, the CO2 concentration time series feature vector and the soil moisture time series feature vector.

3. The AI-based greenhouse planting management and control system according to claim 2 is characterized in that: The soil moisture time series feature embedding unit is used to: after matrix multiplication of the soil moisture time series feature vector and the value embedding matrix, add the obtained soil moisture time series embedding matrix and the value offset vector according to position to obtain the soil moisture time series value vector.

4. The AI-based greenhouse planting management and control system according to claim 3 is characterized in that: The soil moisture-atmospheric environment parameter cross-domain query unit is used to: calculate the product between the query vector and the transposed vector of the update row vector to obtain a query-update association matrix, and then divide the query-update association matrix by the square root of the length of the update row vector by position to obtain a query-update modulation matrix; input the query-update modulation matrix into a softmax function to obtain a query-update activation matrix; multiply the query-update activation matrix by the soil moisture time series value vector to obtain the soil moisture-atmospheric environment parameter cross-domain query attention vector.

5. The AI-based greenhouse planting management and control system according to claim 4 is characterized in that: The control instruction generation module is used to: pass the soil moisture cross-domain query matching feature vector through the decoder-based water inlet valve controller to obtain the recommended value of the water inlet valve opening at the next time point; The control instruction is generated based on a comparison between the current opening value of the water inlet valve and the recommended opening value of the water inlet valve at the next time point.

Images